Rumor Detection Based on Conflict and Bot Features

The complexity of rumors makes them spread fast and challenging to detect. Therefore, multi-modal rumor detection become one of the popular solutions. In this work, we integrate conflict features and bot features into the framework of rumor detection, and by utilizing graph convolutional networks to discern the underlying interplay between these factors, which enhances the model’s understanding of the rumor spread process. Specifically, we employ a pre-trained BERT model to capture stance and construct conflict relations, and employ Bot detectors (i.e., Botometer X and Weibo Bot Finder) to obtain bot score confidences, and finally combine these two features to detect rumors. Experiments on two public datasets, i.e., PHEME and Weibo, validate the effectiveness of our proposed method. Our work supplements current rumor detection methods and highlights the important role of conflict and bot features.